As automobiles gain more capabilities and electrification options the computing power to support them is continually growing through high performance computing (HPC). HPC systems are able to solve complex problems by pooling the computing power of multiple nodes in parallel.
Auto Futures talked with top industry insiders to learn how HPC is set to revolutionise automotive design, software, self-driving and security.
“We are seeing a convergence of AI and high performance computing in the automotive industry. HPC enables faster data processing, and permits the development of high-fidelity environments that can simulate vehicle crash tests and wind-tunnel tests,” says Keith Kirkpatrick, principal analyst, Tractica.
Tim Wong, technical marketing for NVIDIA, a maker of HPC systems, explains that HPC is ideal for when you need a lot of computers to figure out things. Many NVIDIA clients start off with workstations and then grow to enterprise or cloud HPC systems.
Something that can take twelve hours on a workstation could take minutes with HPC computers such as rendering a model. An illustration Wong gives is when architects designed the new NVIDIA headquarters. Before the building was built they noticed while looking at the model that the skylights were in places where the sun would be too hot and then the architect changed the design.
When designing vehicles HPC computing opens up new possibilities. For example, if an automaker is designing an autonomous vehicle there is no steering wheel and no need for human-operated brakes. The designer can easily render an autonomous car with emergency braking so that the wheels may stop by turning perpendicular which could be faster and safer, says Wong.
Perhaps most interestingly, HPC speeds up the data-intensive process of training object-recognition, object-avoidance, and response systems used in autonomous vehicles, says Kirkpatrick.
Wong explains how object training works for autonomous driving. The computers process millions of photos of pedestrians, and learns how to identify them. The data is then used by the computing system in autonomous vehicles to recognise pedestrians.
Another use case for HPC is when an automaker discovers a software glitch or software security vulnerability says Vasilis Kapsalis, director of HPC solutions for Verne Global, supplier of HPC cloud solutions in Iceland that offers the added value of low-cost renewable energy, ISO certification and closed-pipe security. Verne Global’s systems are currently being used by a major German automaker for data processing and by aviation companies for design simulations.
“You can find and heal the software problem and see what is needed to update the system using HPC computing, which could take up to six months otherwise in as little as three days,” says Kapsalis who notes that HPC computing can train algorithms in cases when something goes wrong for providers of car insurance, warranty and rental agencies. He also notes that HPC can also be instrumental in understanding facial expressions.
HPC is currently being used for time-sensitive recommendations such as offering suggestions for ordering on Amazon or providing recommendations for another show to watch on Netflix.
Wong says in the future, when cars are connected to smart cities, real-time data for autonomous vehicles and connected car drivers the help of HPC will be very beneficial.
A camera on a traffic light can identify a jaywalker in front of a bus that is blocking the sensors and camera of the autonomous vehicle behind the bus. The connected autonomous vehicle then ‘knows’ not to attempt to move around the bus due to the jaywalker.
Wong imagines a utopian real-time traffic scenario with connected smart city infrastructure. When there is an event at a huge stadium such as sports match or rock/pop star concert, the traffic lights can be timed to allow tens of thousands of fans to arrive and leave the event easily without massive gridlock.
“We predict that the demand for HPC will rise, given that the amount of data generated by autonomous and connected vehicles is projected to increase exponentially over time,” says Kirkpatrick.
“HPC is likely to be a prime enabler of deep learning within the automotive industry, as deep learning involves the continuous collection and processing of new data to drive and refine algorithms,” he adds.